2 * Xen leaves the responsibility for maintaining p2m mappings to the
3 * guests themselves, but it must also access and update the p2m array
4 * during suspend/resume when all the pages are reallocated.
6 * The p2m table is logically a flat array, but we implement it as a
7 * three-level tree to allow the address space to be sparse.
13 * p2m_mid p2m_mid p2m_mid_mfn p2m_mid_mfn
15 * p2m p2m p2m p2m p2m p2m p2m ...
17 * The p2m_mid_mfn pages are mapped by p2m_top_mfn_p.
19 * The p2m_top and p2m_top_mfn levels are limited to 1 page, so the
20 * maximum representable pseudo-physical address space is:
21 * P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE pages
23 * P2M_PER_PAGE depends on the architecture, as a mfn is always
24 * unsigned long (8 bytes on 64-bit, 4 bytes on 32), leading to
25 * 512 and 1024 entries respectively.
27 * In short, these structures contain the Machine Frame Number (MFN) of the PFN.
29 * However not all entries are filled with MFNs. Specifically for all other
30 * leaf entries, or for the top root, or middle one, for which there is a void
31 * entry, we assume it is "missing". So (for example)
32 * pfn_to_mfn(0x90909090)=INVALID_P2M_ENTRY.
34 * We also have the possibility of setting 1-1 mappings on certain regions, so
36 * pfn_to_mfn(0xc0000)=0xc0000
38 * The benefit of this is, that we can assume for non-RAM regions (think
39 * PCI BARs, or ACPI spaces), we can create mappings easily b/c we
40 * get the PFN value to match the MFN.
42 * For this to work efficiently we have one new page p2m_identity and
43 * allocate (via reserved_brk) any other pages we need to cover the sides
44 * (1GB or 4MB boundary violations). All entries in p2m_identity are set to
45 * INVALID_P2M_ENTRY type (Xen toolstack only recognizes that and MFNs,
46 * no other fancy value).
48 * On lookup we spot that the entry points to p2m_identity and return the
49 * identity value instead of dereferencing and returning INVALID_P2M_ENTRY.
50 * If the entry points to an allocated page, we just proceed as before and
51 * return the PFN. If the PFN has IDENTITY_FRAME_BIT set we unmask that in
52 * appropriate functions (pfn_to_mfn).
54 * The reason for having the IDENTITY_FRAME_BIT instead of just returning the
55 * PFN is that we could find ourselves where pfn_to_mfn(pfn)==pfn for a
56 * non-identity pfn. To protect ourselves against we elect to set (and get) the
57 * IDENTITY_FRAME_BIT on all identity mapped PFNs.
59 * This simplistic diagram is used to explain the more subtle piece of code.
60 * There is also a digram of the P2M at the end that can help.
61 * Imagine your E820 looking as so:
64 * /-------------------+---------\/----\ /----------\ /---+-----\
65 * | System RAM | Sys RAM ||ACPI| | reserved | | Sys RAM |
66 * \-------------------+---------/\----/ \----------/ \---+-----/
69 * [1029MB = 263424 (0x40500), 2001MB = 512256 (0x7D100),
70 * 2048MB = 524288 (0x80000)]
72 * And dom0_mem=max:3GB,1GB is passed in to the guest, meaning memory past 1GB
73 * is actually not present (would have to kick the balloon driver to put it in).
75 * When we are told to set the PFNs for identity mapping (see patch: "xen/setup:
76 * Set identity mapping for non-RAM E820 and E820 gaps.") we pass in the start
77 * of the PFN and the end PFN (263424 and 512256 respectively). The first step
78 * is to reserve_brk a top leaf page if the p2m[1] is missing. The top leaf page
79 * covers 512^2 of page estate (1GB) and in case the start or end PFN is not
80 * aligned on 512^2*PAGE_SIZE (1GB) we loop on aligned 1GB PFNs from start pfn
81 * to end pfn. We reserve_brk top leaf pages if they are missing (means they
82 * point to p2m_mid_missing).
84 * With the E820 example above, 263424 is not 1GB aligned so we allocate a
85 * reserve_brk page which will cover the PFNs estate from 0x40000 to 0x80000.
86 * Each entry in the allocate page is "missing" (points to p2m_missing).
88 * Next stage is to determine if we need to do a more granular boundary check
89 * on the 4MB (or 2MB depending on architecture) off the start and end pfn's.
90 * We check if the start pfn and end pfn violate that boundary check, and if
91 * so reserve_brk a middle (p2m[x][y]) leaf page. This way we have a much finer
92 * granularity of setting which PFNs are missing and which ones are identity.
93 * In our example 263424 and 512256 both fail the check so we reserve_brk two
94 * pages. Populate them with INVALID_P2M_ENTRY (so they both have "missing"
95 * values) and assign them to p2m[1][2] and p2m[1][488] respectively.
97 * At this point we would at minimum reserve_brk one page, but could be up to
98 * three. Each call to set_phys_range_identity has at maximum a three page
99 * cost. If we were to query the P2M at this stage, all those entries from
100 * start PFN through end PFN (so 1029MB -> 2001MB) would return
101 * INVALID_P2M_ENTRY ("missing").
103 * The next step is to walk from the start pfn to the end pfn setting
104 * the IDENTITY_FRAME_BIT on each PFN. This is done in set_phys_range_identity.
105 * If we find that the middle leaf is pointing to p2m_missing we can swap it
106 * over to p2m_identity - this way covering 4MB (or 2MB) PFN space. At this
107 * point we do not need to worry about boundary aligment (so no need to
108 * reserve_brk a middle page, figure out which PFNs are "missing" and which
109 * ones are identity), as that has been done earlier. If we find that the
110 * middle leaf is not occupied by p2m_identity or p2m_missing, we dereference
111 * that page (which covers 512 PFNs) and set the appropriate PFN with
112 * IDENTITY_FRAME_BIT. In our example 263424 and 512256 end up there, and we
113 * set from p2m[1][2][256->511] and p2m[1][488][0->256] with
114 * IDENTITY_FRAME_BIT set.
116 * All other regions that are void (or not filled) either point to p2m_missing
117 * (considered missing) or have the default value of INVALID_P2M_ENTRY (also
118 * considered missing). In our case, p2m[1][2][0->255] and p2m[1][488][257->511]
119 * contain the INVALID_P2M_ENTRY value and are considered "missing."
121 * This is what the p2m ends up looking (for the E820 above) with this
124 * p2m /--------------\
125 * /-----\ | &mfn_list[0],| /-----------------\
126 * | 0 |------>| &mfn_list[1],| /---------------\ | ~0, ~0, .. |
127 * |-----| | ..., ~0, ~0 | | ~0, ~0, [x]---+----->| IDENTITY [@256] |
128 * | 1 |---\ \--------------/ | [p2m_identity]+\ | IDENTITY [@257] |
129 * |-----| \ | [p2m_identity]+\\ | .... |
130 * | 2 |--\ \-------------------->| ... | \\ \----------------/
131 * |-----| \ \---------------/ \\
132 * | 3 |\ \ \\ p2m_identity
133 * |-----| \ \-------------------->/---------------\ /-----------------\
134 * | .. +->+ | [p2m_identity]+-->| ~0, ~0, ~0, ... |
135 * \-----/ / | [p2m_identity]+-->| ..., ~0 |
136 * / /---------------\ | .... | \-----------------/
137 * / | IDENTITY[@0] | /-+-[x], ~0, ~0.. |
138 * / | IDENTITY[@256]|<----/ \---------------/
140 * | \---------------/
142 * p2m_missing p2m_missing
143 * /------------------\ /------------\
144 * | [p2m_mid_missing]+---->| ~0, ~0, ~0 |
145 * | [p2m_mid_missing]+---->| ..., ~0 |
146 * \------------------/ \------------/
148 * where ~0 is INVALID_P2M_ENTRY. IDENTITY is (PFN | IDENTITY_BIT)
151 #include <linux/init.h>
152 #include <linux/module.h>
153 #include <linux/list.h>
154 #include <linux/hash.h>
155 #include <linux/sched.h>
156 #include <linux/seq_file.h>
158 #include <asm/cache.h>
159 #include <asm/setup.h>
161 #include <asm/xen/page.h>
162 #include <asm/xen/hypercall.h>
163 #include <asm/xen/hypervisor.h>
164 #include <xen/grant_table.h>
166 #include "multicalls.h"
169 static void __init
m2p_override_init(void);
171 unsigned long xen_max_p2m_pfn __read_mostly
;
173 #define P2M_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
174 #define P2M_MID_PER_PAGE (PAGE_SIZE / sizeof(unsigned long *))
175 #define P2M_TOP_PER_PAGE (PAGE_SIZE / sizeof(unsigned long **))
177 #define MAX_P2M_PFN (P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE)
179 /* Placeholders for holes in the address space */
180 static RESERVE_BRK_ARRAY(unsigned long, p2m_missing
, P2M_PER_PAGE
);
181 static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_missing
, P2M_MID_PER_PAGE
);
182 static RESERVE_BRK_ARRAY(unsigned long, p2m_mid_missing_mfn
, P2M_MID_PER_PAGE
);
184 static RESERVE_BRK_ARRAY(unsigned long **, p2m_top
, P2M_TOP_PER_PAGE
);
185 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn
, P2M_TOP_PER_PAGE
);
186 static RESERVE_BRK_ARRAY(unsigned long *, p2m_top_mfn_p
, P2M_TOP_PER_PAGE
);
188 static RESERVE_BRK_ARRAY(unsigned long, p2m_identity
, P2M_PER_PAGE
);
190 RESERVE_BRK(p2m_mid
, PAGE_SIZE
* (MAX_DOMAIN_PAGES
/ (P2M_PER_PAGE
* P2M_MID_PER_PAGE
)));
191 RESERVE_BRK(p2m_mid_mfn
, PAGE_SIZE
* (MAX_DOMAIN_PAGES
/ (P2M_PER_PAGE
* P2M_MID_PER_PAGE
)));
193 /* We might hit two boundary violations at the start and end, at max each
194 * boundary violation will require three middle nodes. */
195 RESERVE_BRK(p2m_mid_identity
, PAGE_SIZE
* 2 * 3);
197 /* When we populate back during bootup, the amount of pages can vary. The
198 * max we have is seen is 395979, but that does not mean it can't be more.
199 * But some machines can have 3GB I/O holes even. So lets reserve enough
200 * for 4GB of I/O and E820 holes. */
201 RESERVE_BRK(p2m_populated
, PMD_SIZE
* 4);
202 static inline unsigned p2m_top_index(unsigned long pfn
)
204 BUG_ON(pfn
>= MAX_P2M_PFN
);
205 return pfn
/ (P2M_MID_PER_PAGE
* P2M_PER_PAGE
);
208 static inline unsigned p2m_mid_index(unsigned long pfn
)
210 return (pfn
/ P2M_PER_PAGE
) % P2M_MID_PER_PAGE
;
213 static inline unsigned p2m_index(unsigned long pfn
)
215 return pfn
% P2M_PER_PAGE
;
218 static void p2m_top_init(unsigned long ***top
)
222 for (i
= 0; i
< P2M_TOP_PER_PAGE
; i
++)
223 top
[i
] = p2m_mid_missing
;
226 static void p2m_top_mfn_init(unsigned long *top
)
230 for (i
= 0; i
< P2M_TOP_PER_PAGE
; i
++)
231 top
[i
] = virt_to_mfn(p2m_mid_missing_mfn
);
234 static void p2m_top_mfn_p_init(unsigned long **top
)
238 for (i
= 0; i
< P2M_TOP_PER_PAGE
; i
++)
239 top
[i
] = p2m_mid_missing_mfn
;
242 static void p2m_mid_init(unsigned long **mid
)
246 for (i
= 0; i
< P2M_MID_PER_PAGE
; i
++)
247 mid
[i
] = p2m_missing
;
250 static void p2m_mid_mfn_init(unsigned long *mid
)
254 for (i
= 0; i
< P2M_MID_PER_PAGE
; i
++)
255 mid
[i
] = virt_to_mfn(p2m_missing
);
258 static void p2m_init(unsigned long *p2m
)
262 for (i
= 0; i
< P2M_MID_PER_PAGE
; i
++)
263 p2m
[i
] = INVALID_P2M_ENTRY
;
267 * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
269 * This is called both at boot time, and after resuming from suspend:
270 * - At boot time we're called very early, and must use extend_brk()
271 * to allocate memory.
273 * - After resume we're called from within stop_machine, but the mfn
274 * tree should alreay be completely allocated.
276 void __ref
xen_build_mfn_list_list(void)
280 /* Pre-initialize p2m_top_mfn to be completely missing */
281 if (p2m_top_mfn
== NULL
) {
282 p2m_mid_missing_mfn
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
283 p2m_mid_mfn_init(p2m_mid_missing_mfn
);
285 p2m_top_mfn_p
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
286 p2m_top_mfn_p_init(p2m_top_mfn_p
);
288 p2m_top_mfn
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
289 p2m_top_mfn_init(p2m_top_mfn
);
291 /* Reinitialise, mfn's all change after migration */
292 p2m_mid_mfn_init(p2m_mid_missing_mfn
);
295 for (pfn
= 0; pfn
< xen_max_p2m_pfn
; pfn
+= P2M_PER_PAGE
) {
296 unsigned topidx
= p2m_top_index(pfn
);
297 unsigned mididx
= p2m_mid_index(pfn
);
299 unsigned long *mid_mfn_p
;
301 mid
= p2m_top
[topidx
];
302 mid_mfn_p
= p2m_top_mfn_p
[topidx
];
304 /* Don't bother allocating any mfn mid levels if
305 * they're just missing, just update the stored mfn,
306 * since all could have changed over a migrate.
308 if (mid
== p2m_mid_missing
) {
310 BUG_ON(mid_mfn_p
!= p2m_mid_missing_mfn
);
311 p2m_top_mfn
[topidx
] = virt_to_mfn(p2m_mid_missing_mfn
);
312 pfn
+= (P2M_MID_PER_PAGE
- 1) * P2M_PER_PAGE
;
316 if (mid_mfn_p
== p2m_mid_missing_mfn
) {
318 * XXX boot-time only! We should never find
319 * missing parts of the mfn tree after
320 * runtime. extend_brk() will BUG if we call
323 mid_mfn_p
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
324 p2m_mid_mfn_init(mid_mfn_p
);
326 p2m_top_mfn_p
[topidx
] = mid_mfn_p
;
329 p2m_top_mfn
[topidx
] = virt_to_mfn(mid_mfn_p
);
330 mid_mfn_p
[mididx
] = virt_to_mfn(mid
[mididx
]);
334 void xen_setup_mfn_list_list(void)
336 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
338 HYPERVISOR_shared_info
->arch
.pfn_to_mfn_frame_list_list
=
339 virt_to_mfn(p2m_top_mfn
);
340 HYPERVISOR_shared_info
->arch
.max_pfn
= xen_max_p2m_pfn
;
343 /* Set up p2m_top to point to the domain-builder provided p2m pages */
344 void __init
xen_build_dynamic_phys_to_machine(void)
346 unsigned long *mfn_list
= (unsigned long *)xen_start_info
->mfn_list
;
347 unsigned long max_pfn
= min(MAX_DOMAIN_PAGES
, xen_start_info
->nr_pages
);
350 xen_max_p2m_pfn
= max_pfn
;
352 p2m_missing
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
353 p2m_init(p2m_missing
);
355 p2m_mid_missing
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
356 p2m_mid_init(p2m_mid_missing
);
358 p2m_top
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
359 p2m_top_init(p2m_top
);
361 p2m_identity
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
362 p2m_init(p2m_identity
);
365 * The domain builder gives us a pre-constructed p2m array in
366 * mfn_list for all the pages initially given to us, so we just
367 * need to graft that into our tree structure.
369 for (pfn
= 0; pfn
< max_pfn
; pfn
+= P2M_PER_PAGE
) {
370 unsigned topidx
= p2m_top_index(pfn
);
371 unsigned mididx
= p2m_mid_index(pfn
);
373 if (p2m_top
[topidx
] == p2m_mid_missing
) {
374 unsigned long **mid
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
377 p2m_top
[topidx
] = mid
;
381 * As long as the mfn_list has enough entries to completely
382 * fill a p2m page, pointing into the array is ok. But if
383 * not the entries beyond the last pfn will be undefined.
385 if (unlikely(pfn
+ P2M_PER_PAGE
> max_pfn
)) {
386 unsigned long p2midx
;
388 p2midx
= max_pfn
% P2M_PER_PAGE
;
389 for ( ; p2midx
< P2M_PER_PAGE
; p2midx
++)
390 mfn_list
[pfn
+ p2midx
] = INVALID_P2M_ENTRY
;
392 p2m_top
[topidx
][mididx
] = &mfn_list
[pfn
];
398 unsigned long get_phys_to_machine(unsigned long pfn
)
400 unsigned topidx
, mididx
, idx
;
402 if (unlikely(pfn
>= MAX_P2M_PFN
))
403 return INVALID_P2M_ENTRY
;
405 topidx
= p2m_top_index(pfn
);
406 mididx
= p2m_mid_index(pfn
);
407 idx
= p2m_index(pfn
);
410 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
411 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
414 if (p2m_top
[topidx
][mididx
] == p2m_identity
)
415 return IDENTITY_FRAME(pfn
);
417 return p2m_top
[topidx
][mididx
][idx
];
419 EXPORT_SYMBOL_GPL(get_phys_to_machine
);
421 static void *alloc_p2m_page(void)
423 return (void *)__get_free_page(GFP_KERNEL
| __GFP_REPEAT
);
426 static void free_p2m_page(void *p
)
428 free_page((unsigned long)p
);
432 * Fully allocate the p2m structure for a given pfn. We need to check
433 * that both the top and mid levels are allocated, and make sure the
434 * parallel mfn tree is kept in sync. We may race with other cpus, so
435 * the new pages are installed with cmpxchg; if we lose the race then
436 * simply free the page we allocated and use the one that's there.
438 static bool alloc_p2m(unsigned long pfn
)
440 unsigned topidx
, mididx
;
441 unsigned long ***top_p
, **mid
;
442 unsigned long *top_mfn_p
, *mid_mfn
;
444 topidx
= p2m_top_index(pfn
);
445 mididx
= p2m_mid_index(pfn
);
447 top_p
= &p2m_top
[topidx
];
450 if (mid
== p2m_mid_missing
) {
451 /* Mid level is missing, allocate a new one */
452 mid
= alloc_p2m_page();
458 if (cmpxchg(top_p
, p2m_mid_missing
, mid
) != p2m_mid_missing
)
462 top_mfn_p
= &p2m_top_mfn
[topidx
];
463 mid_mfn
= p2m_top_mfn_p
[topidx
];
465 BUG_ON(virt_to_mfn(mid_mfn
) != *top_mfn_p
);
467 if (mid_mfn
== p2m_mid_missing_mfn
) {
468 /* Separately check the mid mfn level */
469 unsigned long missing_mfn
;
470 unsigned long mid_mfn_mfn
;
472 mid_mfn
= alloc_p2m_page();
476 p2m_mid_mfn_init(mid_mfn
);
478 missing_mfn
= virt_to_mfn(p2m_mid_missing_mfn
);
479 mid_mfn_mfn
= virt_to_mfn(mid_mfn
);
480 if (cmpxchg(top_mfn_p
, missing_mfn
, mid_mfn_mfn
) != missing_mfn
)
481 free_p2m_page(mid_mfn
);
483 p2m_top_mfn_p
[topidx
] = mid_mfn
;
486 if (p2m_top
[topidx
][mididx
] == p2m_identity
||
487 p2m_top
[topidx
][mididx
] == p2m_missing
) {
488 /* p2m leaf page is missing */
490 unsigned long *p2m_orig
= p2m_top
[topidx
][mididx
];
492 p2m
= alloc_p2m_page();
498 if (cmpxchg(&mid
[mididx
], p2m_orig
, p2m
) != p2m_orig
)
501 mid_mfn
[mididx
] = virt_to_mfn(p2m
);
507 static bool __init
early_alloc_p2m_middle(unsigned long pfn
, bool check_boundary
)
509 unsigned topidx
, mididx
, idx
;
511 unsigned long *mid_mfn_p
;
513 topidx
= p2m_top_index(pfn
);
514 mididx
= p2m_mid_index(pfn
);
515 idx
= p2m_index(pfn
);
517 /* Pfff.. No boundary cross-over, lets get out. */
518 if (!idx
&& check_boundary
)
521 WARN(p2m_top
[topidx
][mididx
] == p2m_identity
,
522 "P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n",
526 * Could be done by xen_build_dynamic_phys_to_machine..
528 if (p2m_top
[topidx
][mididx
] != p2m_missing
)
531 /* Boundary cross-over for the edges: */
532 p2m
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
536 p2m_top
[topidx
][mididx
] = p2m
;
538 /* For save/restore we need to MFN of the P2M saved */
540 mid_mfn_p
= p2m_top_mfn_p
[topidx
];
541 WARN(mid_mfn_p
[mididx
] != virt_to_mfn(p2m_missing
),
542 "P2M_TOP_P[%d][%d] != MFN of p2m_missing!\n",
544 mid_mfn_p
[mididx
] = virt_to_mfn(p2m
);
549 static bool __init
early_alloc_p2m(unsigned long pfn
)
551 unsigned topidx
= p2m_top_index(pfn
);
552 unsigned long *mid_mfn_p
;
555 mid
= p2m_top
[topidx
];
556 mid_mfn_p
= p2m_top_mfn_p
[topidx
];
557 if (mid
== p2m_mid_missing
) {
558 mid
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
562 p2m_top
[topidx
] = mid
;
564 BUG_ON(mid_mfn_p
!= p2m_mid_missing_mfn
);
566 /* And the save/restore P2M tables.. */
567 if (mid_mfn_p
== p2m_mid_missing_mfn
) {
568 mid_mfn_p
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
569 p2m_mid_mfn_init(mid_mfn_p
);
571 p2m_top_mfn_p
[topidx
] = mid_mfn_p
;
572 p2m_top_mfn
[topidx
] = virt_to_mfn(mid_mfn_p
);
573 /* Note: we don't set mid_mfn_p[midix] here,
574 * look in early_alloc_p2m_middle */
578 bool __init
early_set_phys_to_machine(unsigned long pfn
, unsigned long mfn
)
580 if (unlikely(!__set_phys_to_machine(pfn
, mfn
))) {
581 if (!early_alloc_p2m(pfn
))
584 if (!early_alloc_p2m_middle(pfn
, false /* boundary crossover OK!*/))
587 if (!__set_phys_to_machine(pfn
, mfn
))
593 unsigned long __init
set_phys_range_identity(unsigned long pfn_s
,
598 if (unlikely(pfn_s
>= MAX_P2M_PFN
|| pfn_e
>= MAX_P2M_PFN
))
601 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap
)))
602 return pfn_e
- pfn_s
;
607 for (pfn
= (pfn_s
& ~(P2M_MID_PER_PAGE
* P2M_PER_PAGE
- 1));
608 pfn
< ALIGN(pfn_e
, (P2M_MID_PER_PAGE
* P2M_PER_PAGE
));
609 pfn
+= P2M_MID_PER_PAGE
* P2M_PER_PAGE
)
611 WARN_ON(!early_alloc_p2m(pfn
));
614 early_alloc_p2m_middle(pfn_s
, true);
615 early_alloc_p2m_middle(pfn_e
, true);
617 for (pfn
= pfn_s
; pfn
< pfn_e
; pfn
++)
618 if (!__set_phys_to_machine(pfn
, IDENTITY_FRAME(pfn
)))
621 if (!WARN((pfn
- pfn_s
) != (pfn_e
- pfn_s
),
622 "Identity mapping failed. We are %ld short of 1-1 mappings!\n",
623 (pfn_e
- pfn_s
) - (pfn
- pfn_s
)))
624 printk(KERN_DEBUG
"1-1 mapping on %lx->%lx\n", pfn_s
, pfn
);
629 /* Try to install p2m mapping; fail if intermediate bits missing */
630 bool __set_phys_to_machine(unsigned long pfn
, unsigned long mfn
)
632 unsigned topidx
, mididx
, idx
;
634 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap
))) {
635 BUG_ON(pfn
!= mfn
&& mfn
!= INVALID_P2M_ENTRY
);
638 if (unlikely(pfn
>= MAX_P2M_PFN
)) {
639 BUG_ON(mfn
!= INVALID_P2M_ENTRY
);
643 topidx
= p2m_top_index(pfn
);
644 mididx
= p2m_mid_index(pfn
);
645 idx
= p2m_index(pfn
);
647 /* For sparse holes were the p2m leaf has real PFN along with
648 * PCI holes, stick in the PFN as the MFN value.
650 if (mfn
!= INVALID_P2M_ENTRY
&& (mfn
& IDENTITY_FRAME_BIT
)) {
651 if (p2m_top
[topidx
][mididx
] == p2m_identity
)
654 /* Swap over from MISSING to IDENTITY if needed. */
655 if (p2m_top
[topidx
][mididx
] == p2m_missing
) {
656 WARN_ON(cmpxchg(&p2m_top
[topidx
][mididx
], p2m_missing
,
657 p2m_identity
) != p2m_missing
);
662 if (p2m_top
[topidx
][mididx
] == p2m_missing
)
663 return mfn
== INVALID_P2M_ENTRY
;
665 p2m_top
[topidx
][mididx
][idx
] = mfn
;
670 bool set_phys_to_machine(unsigned long pfn
, unsigned long mfn
)
672 if (unlikely(!__set_phys_to_machine(pfn
, mfn
))) {
676 if (!__set_phys_to_machine(pfn
, mfn
))
683 #define M2P_OVERRIDE_HASH_SHIFT 10
684 #define M2P_OVERRIDE_HASH (1 << M2P_OVERRIDE_HASH_SHIFT)
686 static RESERVE_BRK_ARRAY(struct list_head
, m2p_overrides
, M2P_OVERRIDE_HASH
);
687 static DEFINE_SPINLOCK(m2p_override_lock
);
689 static void __init
m2p_override_init(void)
693 m2p_overrides
= extend_brk(sizeof(*m2p_overrides
) * M2P_OVERRIDE_HASH
,
694 sizeof(unsigned long));
696 for (i
= 0; i
< M2P_OVERRIDE_HASH
; i
++)
697 INIT_LIST_HEAD(&m2p_overrides
[i
]);
700 static unsigned long mfn_hash(unsigned long mfn
)
702 return hash_long(mfn
, M2P_OVERRIDE_HASH_SHIFT
);
705 /* Add an MFN override for a particular page */
706 int m2p_add_override(unsigned long mfn
, struct page
*page
,
707 struct gnttab_map_grant_ref
*kmap_op
)
711 unsigned long uninitialized_var(address
);
716 pfn
= page_to_pfn(page
);
717 if (!PageHighMem(page
)) {
718 address
= (unsigned long)__va(pfn
<< PAGE_SHIFT
);
719 ptep
= lookup_address(address
, &level
);
720 if (WARN(ptep
== NULL
|| level
!= PG_LEVEL_4K
,
721 "m2p_add_override: pfn %lx not mapped", pfn
))
724 WARN_ON(PagePrivate(page
));
725 SetPagePrivate(page
);
726 set_page_private(page
, mfn
);
727 page
->index
= pfn_to_mfn(pfn
);
729 if (unlikely(!set_phys_to_machine(pfn
, FOREIGN_FRAME(mfn
))))
732 if (kmap_op
!= NULL
) {
733 if (!PageHighMem(page
)) {
734 struct multicall_space mcs
=
735 xen_mc_entry(sizeof(*kmap_op
));
737 MULTI_grant_table_op(mcs
.mc
,
738 GNTTABOP_map_grant_ref
, kmap_op
, 1);
740 xen_mc_issue(PARAVIRT_LAZY_MMU
);
742 /* let's use dev_bus_addr to record the old mfn instead */
743 kmap_op
->dev_bus_addr
= page
->index
;
744 page
->index
= (unsigned long) kmap_op
;
746 spin_lock_irqsave(&m2p_override_lock
, flags
);
747 list_add(&page
->lru
, &m2p_overrides
[mfn_hash(mfn
)]);
748 spin_unlock_irqrestore(&m2p_override_lock
, flags
);
750 /* p2m(m2p(mfn)) == mfn: the mfn is already present somewhere in
751 * this domain. Set the FOREIGN_FRAME_BIT in the p2m for the other
752 * pfn so that the following mfn_to_pfn(mfn) calls will return the
753 * pfn from the m2p_override (the backend pfn) instead.
754 * We need to do this because the pages shared by the frontend
755 * (xen-blkfront) can be already locked (lock_page, called by
756 * do_read_cache_page); when the userspace backend tries to use them
757 * with direct_IO, mfn_to_pfn returns the pfn of the frontend, so
758 * do_blockdev_direct_IO is going to try to lock the same pages
759 * again resulting in a deadlock.
760 * As a side effect get_user_pages_fast might not be safe on the
761 * frontend pages while they are being shared with the backend,
762 * because mfn_to_pfn (that ends up being called by GUPF) will
763 * return the backend pfn rather than the frontend pfn. */
764 ret
= __get_user(pfn
, &machine_to_phys_mapping
[mfn
]);
765 if (ret
== 0 && get_phys_to_machine(pfn
) == mfn
)
766 set_phys_to_machine(pfn
, FOREIGN_FRAME(mfn
));
770 EXPORT_SYMBOL_GPL(m2p_add_override
);
771 int m2p_remove_override(struct page
*page
, bool clear_pte
)
776 unsigned long uninitialized_var(address
);
781 pfn
= page_to_pfn(page
);
782 mfn
= get_phys_to_machine(pfn
);
783 if (mfn
== INVALID_P2M_ENTRY
|| !(mfn
& FOREIGN_FRAME_BIT
))
786 if (!PageHighMem(page
)) {
787 address
= (unsigned long)__va(pfn
<< PAGE_SHIFT
);
788 ptep
= lookup_address(address
, &level
);
790 if (WARN(ptep
== NULL
|| level
!= PG_LEVEL_4K
,
791 "m2p_remove_override: pfn %lx not mapped", pfn
))
795 spin_lock_irqsave(&m2p_override_lock
, flags
);
796 list_del(&page
->lru
);
797 spin_unlock_irqrestore(&m2p_override_lock
, flags
);
798 WARN_ON(!PagePrivate(page
));
799 ClearPagePrivate(page
);
802 struct gnttab_map_grant_ref
*map_op
=
803 (struct gnttab_map_grant_ref
*) page
->index
;
804 set_phys_to_machine(pfn
, map_op
->dev_bus_addr
);
805 if (!PageHighMem(page
)) {
806 struct multicall_space mcs
;
807 struct gnttab_unmap_grant_ref
*unmap_op
;
810 * It might be that we queued all the m2p grant table
811 * hypercalls in a multicall, then m2p_remove_override
812 * get called before the multicall has actually been
813 * issued. In this case handle is going to -1 because
814 * it hasn't been modified yet.
816 if (map_op
->handle
== -1)
819 * Now if map_op->handle is negative it means that the
820 * hypercall actually returned an error.
822 if (map_op
->handle
== GNTST_general_error
) {
823 printk(KERN_WARNING
"m2p_remove_override: "
824 "pfn %lx mfn %lx, failed to modify kernel mappings",
830 sizeof(struct gnttab_unmap_grant_ref
));
832 unmap_op
->host_addr
= map_op
->host_addr
;
833 unmap_op
->handle
= map_op
->handle
;
834 unmap_op
->dev_bus_addr
= 0;
836 MULTI_grant_table_op(mcs
.mc
,
837 GNTTABOP_unmap_grant_ref
, unmap_op
, 1);
839 xen_mc_issue(PARAVIRT_LAZY_MMU
);
841 set_pte_at(&init_mm
, address
, ptep
,
842 pfn_pte(pfn
, PAGE_KERNEL
));
843 __flush_tlb_single(address
);
844 map_op
->host_addr
= 0;
847 set_phys_to_machine(pfn
, page
->index
);
849 /* p2m(m2p(mfn)) == FOREIGN_FRAME(mfn): the mfn is already present
850 * somewhere in this domain, even before being added to the
851 * m2p_override (see comment above in m2p_add_override).
852 * If there are no other entries in the m2p_override corresponding
853 * to this mfn, then remove the FOREIGN_FRAME_BIT from the p2m for
854 * the original pfn (the one shared by the frontend): the backend
855 * cannot do any IO on this page anymore because it has been
856 * unshared. Removing the FOREIGN_FRAME_BIT from the p2m entry of
857 * the original pfn causes mfn_to_pfn(mfn) to return the frontend
859 mfn
&= ~FOREIGN_FRAME_BIT
;
860 ret
= __get_user(pfn
, &machine_to_phys_mapping
[mfn
]);
861 if (ret
== 0 && get_phys_to_machine(pfn
) == FOREIGN_FRAME(mfn
) &&
862 m2p_find_override(mfn
) == NULL
)
863 set_phys_to_machine(pfn
, mfn
);
867 EXPORT_SYMBOL_GPL(m2p_remove_override
);
869 struct page
*m2p_find_override(unsigned long mfn
)
872 struct list_head
*bucket
= &m2p_overrides
[mfn_hash(mfn
)];
873 struct page
*p
, *ret
;
877 spin_lock_irqsave(&m2p_override_lock
, flags
);
879 list_for_each_entry(p
, bucket
, lru
) {
880 if (page_private(p
) == mfn
) {
886 spin_unlock_irqrestore(&m2p_override_lock
, flags
);
891 unsigned long m2p_find_override_pfn(unsigned long mfn
, unsigned long pfn
)
893 struct page
*p
= m2p_find_override(mfn
);
894 unsigned long ret
= pfn
;
897 ret
= page_to_pfn(p
);
901 EXPORT_SYMBOL_GPL(m2p_find_override_pfn
);
903 #ifdef CONFIG_XEN_DEBUG_FS
904 #include <linux/debugfs.h>
906 static int p2m_dump_show(struct seq_file
*m
, void *v
)
908 static const char * const level_name
[] = { "top", "middle",
909 "entry", "abnormal", "error"};
910 #define TYPE_IDENTITY 0
911 #define TYPE_MISSING 1
913 #define TYPE_UNKNOWN 3
914 static const char * const type_name
[] = {
915 [TYPE_IDENTITY
] = "identity",
916 [TYPE_MISSING
] = "missing",
918 [TYPE_UNKNOWN
] = "abnormal"};
919 unsigned long pfn
, prev_pfn_type
= 0, prev_pfn_level
= 0;
920 unsigned int uninitialized_var(prev_level
);
921 unsigned int uninitialized_var(prev_type
);
926 for (pfn
= 0; pfn
< MAX_DOMAIN_PAGES
; pfn
++) {
927 unsigned topidx
= p2m_top_index(pfn
);
928 unsigned mididx
= p2m_mid_index(pfn
);
929 unsigned idx
= p2m_index(pfn
);
934 if (p2m_top
[topidx
] == p2m_mid_missing
) {
935 lvl
= 0; type
= TYPE_MISSING
;
936 } else if (p2m_top
[topidx
] == NULL
) {
937 lvl
= 0; type
= TYPE_UNKNOWN
;
938 } else if (p2m_top
[topidx
][mididx
] == NULL
) {
939 lvl
= 1; type
= TYPE_UNKNOWN
;
940 } else if (p2m_top
[topidx
][mididx
] == p2m_identity
) {
941 lvl
= 1; type
= TYPE_IDENTITY
;
942 } else if (p2m_top
[topidx
][mididx
] == p2m_missing
) {
943 lvl
= 1; type
= TYPE_MISSING
;
944 } else if (p2m_top
[topidx
][mididx
][idx
] == 0) {
945 lvl
= 2; type
= TYPE_UNKNOWN
;
946 } else if (p2m_top
[topidx
][mididx
][idx
] == IDENTITY_FRAME(pfn
)) {
947 lvl
= 2; type
= TYPE_IDENTITY
;
948 } else if (p2m_top
[topidx
][mididx
][idx
] == INVALID_P2M_ENTRY
) {
949 lvl
= 2; type
= TYPE_MISSING
;
950 } else if (p2m_top
[topidx
][mididx
][idx
] == pfn
) {
951 lvl
= 2; type
= TYPE_PFN
;
952 } else if (p2m_top
[topidx
][mididx
][idx
] != pfn
) {
953 lvl
= 2; type
= TYPE_PFN
;
959 if (pfn
== MAX_DOMAIN_PAGES
-1) {
963 if (prev_type
!= type
) {
964 seq_printf(m
, " [0x%lx->0x%lx] %s\n",
965 prev_pfn_type
, pfn
, type_name
[prev_type
]);
969 if (prev_level
!= lvl
) {
970 seq_printf(m
, " [0x%lx->0x%lx] level %s\n",
971 prev_pfn_level
, pfn
, level_name
[prev_level
]);
972 prev_pfn_level
= pfn
;
983 static int p2m_dump_open(struct inode
*inode
, struct file
*filp
)
985 return single_open(filp
, p2m_dump_show
, NULL
);
988 static const struct file_operations p2m_dump_fops
= {
989 .open
= p2m_dump_open
,
992 .release
= single_release
,
995 static struct dentry
*d_mmu_debug
;
997 static int __init
xen_p2m_debugfs(void)
999 struct dentry
*d_xen
= xen_init_debugfs();
1004 d_mmu_debug
= debugfs_create_dir("mmu", d_xen
);
1006 debugfs_create_file("p2m", 0600, d_mmu_debug
, NULL
, &p2m_dump_fops
);
1009 fs_initcall(xen_p2m_debugfs
);
1010 #endif /* CONFIG_XEN_DEBUG_FS */